Let's be Fwends #133:

You Really Shouldn't Sit That Much

Do not go gentle into that good night.
Rage, rage against the dying of the light.

~ Dylan Thomas

Are Large Language Models just the next milestone on a road to disappointment trying to understand what 'intelligence' is? What's the maximum duration you should sit each day? Who keeps the Internet running and how fast will China outspend the rest of the world on renewable energy? Welcome to Let's be Fwends #133.

How Much Time You Should Spend Standing (vs. Sitting)

If you're like me, you're spending most of your (working) days sitting in front of a computer. If you're a bit more like me, you bought the "sitting is the new smoking" hype and got yourself a standing desk so you can alternate between standing and sitting while at work. And if you're even more like me (but not in a spooky way, but more in a well-this-is-actually-pretty-common-for-standing-desk-owners-way), you don't really use your standing desk - except for sitting in front of it.

To, it was always a bit unclear how to properly use my standing desk - should I go full standing (and then evolving to walking on a treadmill like Stephen Wolfram?), or alternate standing and sitting in 20 minute bouts? That's awkward in meetings, and I easily lose track.

Well, thanks to new research, I now know that you shouldn't sit for more than six hours a day.

With the exercise-part of the research findings covered, I consider setting myself a daily "sitting budget" that I keep track of. Maybe using my standing desk "the other way round" (limiting sitting time, instead of maximising standing time) will work better.

How LLMs Work Explained - Without Math

I've covered the inner workings of LLMs before, but if you want something simpler, here's a great explainer on LLMs and how they actually work without the math.

During my studies, one of my professors pointed out that in our search of Artificial Intelligence, we keep moving the goal posts: First we thought that being able to play chess is a clear sign of intelligence. Then, Deep Blue came along and brute-forced the problem. After that, we thought checkers would be a better test. The game is so much more complex than chess that the algorithms used to beat humans at chess could not be used. Then we found out that you can apply heuristics to play the perfect checkers game. Then, we resorted back to the Turing Test: When a computer can trick a human interviewer to think it is human, too, we shall consider this computer to be "intelligent". But now we know (and the article above explains why and how) that LLMs are nothing more than brute-force machines, just at an unprecedented scale in designing them, constructing them, training them and running them. Nothing in the workings of an LLM points to anything that we would consider 'intelligent'.

Since all our approaches to defining "intelligence" can be solved by brute-force - is intelligence actually nothing more than the computation of basic information on a big scale? Is it an emergent property of a sufficiently complex system?

Or does that "special magic sauce" philosophers insisted on for millennia really exist, and we're unable to accurately describe what makes a mind a mind, exactly because of that?

Renewable Energy in China

If you're talking about climate change and green energy, one of the most often heard counter-arguments (at least here in Europe) is that as long as China (one of the heaviest polluters when it comes to CO2 emissions) doesn't change its energy-generation policy, it won't matter if smaller countries switch to renewable energy sources.

This is - of course - an obvious attempt to avoid any change yourself: As long as this other person over there is not changing what they're doing, I won't change what I'm doing. Conveniently, everyone is pointing fingers at everyone else so everyone can keep doing what they're doing.

Except - it's not true.

China is switching to green energies on an unprecedented scale:

"In 2022, China installed roughly as much solar capacity as the rest of the world combined, then doubled additional solar in 2023."

Source

Additionally, it is abandoning nuclear energy as the backbone of their low-emission energy grid, switching to renewables like solar- and wind-power, simply because building nuclear power plants takes too long, is too expensive and doesn't scale:

"(E)ven in China, (...) nuclear energy cannot compete with renewable energy to deliver low emission electricity generation at the deployment rates needed to meet mid-century emission targets."

Source

So, the next time someone tells you that the next car they'll be buying will be another gas-powered SUV because China, you can tell them that while China still has some ground to make up, they're done walking and started running.

The People Who Make Sure the Internet Continues to Work

The Verge wrote an amazing profile on the people who keep the Internet running and how hard that job is

If you think "keep the Internet running", think "boats and underwater cables", not "attics and consoles".

Rage, Rage Against the Dying of the Light

Michael Sheen performing Do not go gentle into this good night is a masterpiece.

That's it for this edition of Let's be Fwends. You can now sit down again. Don't overdo it. 🪑